9 September 2016

Motivation

Traditionally, evaluating the outcome (effect) of behaviour change interventions in household energy studies have been given far more importance than studying the underlying processes that is assumed to have led to the outcome.

A dynamic approach to psychology

Traditional Dynamic
cross sectional data intensive longitudinal data
static relationships underlying processes
(mostly) confirmatory exploratory and confirmatory
multiple regressions time series and dynamic models

Exploratory research

SGR Buurkracht
N = 100 (households) N = 142 (neighbourhoods)
t = 2 years (2013 - 2015) t = 2 years (starting Dec 2015)
feedback, demand side management feedback, adoption of PV
baseline (3 months) baseline (6 months)
non equivalent control group non equivalent control group

Data structure

The Data Box (Catell, 1952)

Intensive longitudinal data (ILD)

Psychological determinants of household energy consumption

Intensive longitudinal data (ILD)

Linear Time Series Models


\(y_n = c + \sum_{i=1}^{p}\phi_{i}y_{n-i} + \sum_{i=1}^{p}\theta_{i}\epsilon_{n-i} + \epsilon_n\)

ARIMA Model Estimation

ARIMA Model Diagnostics

Hidden Markov Models (HMM)


\(P(Y,Z) = P(z_{1})\left[\prod_{n=2}^{N} P(z_{n}|z_{n-1})\right] \prod_{n=1}^{N}P(y_{n}|z_{n})\)

HMM Inference

Results

Model AIC BIC
ARIMA(1,0,2) 312261.9 312302.6
HMM (k=3) 301502.2 301618.7

When comparing models fitted by maximum likelihood to the same data, the smaller the AIC or BIC, the better the fit.

Summary and future work

  • to truly understand the impact of behaviour change programs, we need to study processes
  • to study underlying processes, we require ILD
  • to model ILD, we require dynamic models
  • behaviour charecterized by discrete regimes, each of which has approximately linear dynamics
  • Bayesian hierarchical modelling (within vs between person variations)
  • system dynamics as model-based theory building

Thank you